Search This Blog

Sunday, 28 August 2016

Really proud of my students -- final projects

Over the next few weeks I'm going to do some short blog posts about each of the final projects my students did in their data science course.

One of the reasons this blog has been a bit quieter than usual these last few months is that I was teaching a Data Science class at General Assembly, which was rewarding but rather exhausting.

Some observations:
  • GA is busy and dynamic. I remember back in the late 1990s at HP when every company was deploying SAP on HP-UX to avoid Y2K problems: there were classes constantly; you might discover that the class you were teaching was going to be held in the boardroom using some workstations borrowed from another city. GA was like that: every room packed from early morning until late at night.
  • No-one in the class had a job as a data scientist at the beginning of the course, but there was a lot of movement within 10 weeks: job changes, promotions, new career directions. The only time in my teaching career where I saw the same wow-this-person-is-trained-now-let's-poach-them was in the early days of the Peregrine -> Service Manager transition.
  • The course is mainly about machine learning but there is flexibility for the instructor to add in a few other relevant topics based on what the students want. Right now, Natural Language Processing is white-hot. Several students did some serious NLP / NLU projects. The opportunities for people who have skills in this area are very, very good.
  • Computer vision is an area where there is a lot of interest as well.
I'll be teaching the first part of the Data Science immersive (a full-time course instead of a night-time part-time one) starting in September; please sign up with GA if you are interested.

I suspect by the time I've finished blogging about my past students' projects that there will be a new round of student projects to cover, so this might become a bit more of a feature on my blog.

Tuesday, 23 August 2016

Automate the installation of a Windows DataProtector client

A client today wanted to push the DataProtector agent from SCCM / System Center 2012 instead of from Data Protector. It's not that difficult, but I couldn't find the command-line setup documented anywhere.

You will need to run (as an administrator):

  net use r: \\\Omniback
  cd \x8664
  msiexec /i "Data Protector A.09.00.msi" /passive INSTALLATIONTYPE=Client ADDLOCAL=core,da,autodr
  net use /delete r:

Obviously, substitute with your install server, and if R: is already allocated, use something else instead.

Then, trigger the following command on your cell manager:

 omnicc -import_host clientname

Replace clientname with the name of the client.

Script this as appropriate (e.g. after the operating system has booted) in order to have an unattended installation.

Friday, 5 August 2016

Data Protector CRS operation cannot be performed in full-screen mode

Today's head-scratcher: after upgrading to 9.07 on a Windows cell manager, the CRS service won't start.

Eventvwr says something even weirder:

The Data Protector CRS service terminated with service-specific error The requested operation cannot be performed in full-screen mode.. 

I was in full screen mode at the time, but it still wouldn't start even when I minimised my RDP session. For my own sanity, I was glad of this.

Trawling through Daniel Braun's blog, I saw some comments there that it could be related to anti-virus software. Nope, not that either.

The debug.log said something a little bit more believable:

[SmCreateTable] MapViewOfFile(size:17505216) failed, error=[5] Access is denied.

I discovered that I could reliably get that message added that message every time I tried to start the CRS. But what is actually being denied?

So I ran omnisv start -debug 1-500 crm-vexatious.txt

I then had a 160KB file created in C:\programdata\omniback\tmp that began with OB2DBG, ended with crm-vexatious.txt and had CRM in the filename. Good: at least it gets far enough that it can create debug messages.

Scrollling right to the bottom of it, there it was:

Code is:1007  SystemErr: [5] Access is denied
************************   DEFAULT ERROR REPORT   ***************
[Critical] From [email protected] "" Time: 5/8/2016 1:00:33PM
Unable to allocate shared memory: Unknown internal error.

Internally, the function to return a shared memory segement presumably encodes something as 1007; CRS then exits with that code (which is the standard Windows error code for "can't be performed in full-screen mode").

There aren't many reasons for a shared memory allocation to fail. In fact, the only one I can think of that could be relevant here is if the segment already exists. I thought about figuring out what the equivalent to ipcrm is on Windows, gave up and rebooted the box.

And it came up perfectly. Funnily enough, if I had had no idea what I was doing, I would have just bounced the box to see if it would have fixed it, and saved myself a headache and some stress wondering what was going on. Ignorance would have been bliss.